M Atif Ali

Product Manager

<p><strong>Note</strong>: AI Bridge is now <em>AI Gateway</em>, and Agent Boundaries is now <em>Agent Firewall</em>. Both remain part of the Coder AI Governance solution with no functional changes.</p>

<p>The shift from &quot;generative&quot; to &quot;agentic&quot; AI creates friction between developer velocity and enterprise security. We&#39;re way past chatbots. Developers are deploying autonomous agents like Anthropic&#39;s Claude Code that navigate file systems, execute terminal commands, and refactor entire services in a loop.</p>

<p>For developers, this is a superpower. For platform teams, it&#39;s a governance nightmare. According to Gartner, over <a href="https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027">40% of agentic AI projects</a> will be scrapped by 2027 because enterprises can&#39;t govern it safely, not necessarily because the technology doesn&#39;t work. The problem isn&#39;t the agents themselves. It&#39;s where they run and how they access resources.</p>

<p>How do you safely enable tools designed to have &quot;hands&quot; on the command line? Most teams are forced into a &quot;shadow AI&quot; workflow:</p>

<ul>
<li><strong>Key sprawl:</strong> Developers generate personal API keys (<code>sk-...</code>) and store them in plaintext <code>.env</code> files or shell histories on their laptops.</li>
<li><strong>MCP tool sprawl:</strong> Developers configure MCP servers locally without oversight, leaving platform teams blind to what tools are accessing company resources or what data is being exposed.</li>
<li><strong>Context leakage:</strong> To be useful, agents need context. Developers often paste proprietary schemas or database dumps into the agent&#39;s context window, bypassing data loss prevention (DLP) controls.</li>
</ul>

<p>Blocking these tools isn&#39;t an option. They&#39;re too valuable. The solution isn&#39;t to restrict them—it&#39;s to rethink where they run.</p>


<h2>The workspace should be the agent&#39;s home</h2>

<p>An autonomous coding agent shouldn&#39;t run on a developer&#39;s laptop, constrained by local networking rules and unmanaged secrets. It should run in the cloud within the same secure boundary as the workload it modifies.</p>

<p>When agents run locally, every security control becomes reactive: DLP scanning after the fact, key rotation after compromise, audit logs that can&#39;t prove a negative. But when agents run inside governed infrastructure, security becomes architectural and built into the environment rather than bolted on after.</p>

<p>The question isn&#39;t &quot;how do we monitor what developers do with AI?&quot; It&#39;s &quot;how do we make AI work the way enterprise systems already work?&quot;</p>

<p>In this model, the cloud development environment acts as the control plane or &quot;immune system&quot; wrapping around the agent, providing identity, authorized tools, and secure connectivity. Enterprise AI adoption hinges on <strong>centralized federation</strong>, where the platform injects authentication and tools just-in-time, and the agent never holds the keys to the kingdom.</p>


<h2>Meet AI Bridge</h2>

<p>Facing all this, enterprises need a centralized control plane for agents that provides governance, not just monitoring. That&#39;s exactly what we built with AI Bridge. </p>

<p>AI Bridge acts as middleware between your developers&#39; agents and external LLM providers, running as an in-process daemon inside the Coder control plane. This effectively solves the Shadow AI problem at the infrastructure level:</p>


AI Bridge

<h3>Identity-aware LLM access with AI Bridge</h3>

<ul>
<li><strong>No more local keys:</strong> A user&#39;s Coder API key authenticates the AI agent with AI Bridge. An admin can configure this at the template level to be automatically injected as an environment variable (e.g., <code>ANTHROPIC_API_KEY</code>). The agent talks to AI Bridge on a custom base URL set as <code>ANTHROPIC_BASE_URL=https://coder.example.com/api/v2/aibridge/anthropic</code>, and AI Bridge talks to the LLM provider using your existing enterprise credentials.</li>
<li><strong>User attribution:</strong> Traffic flows through AI Bridge, so every prompt and request is tagged with each user&#39;s identity. You get a granular audit trail showing who used which model and for what purpose.</li>
<li><strong>Spend tracking:</strong> Monitor token spend and costs per user, team, or organization for chargeback. Set up alerts for abnormal usage patterns. Identify power users and least engaged users.</li>
<li><strong>MCP tools auditing:</strong> Gain comprehensive visibility into which specific MCP tools your agents and developers are accessing, along with timestamps of when each access occurred, providing a complete audit trail for security and compliance purposes.</li>
</ul>


<h2>How it works in the real world</h2>

<p>Consider a team adopting Claude Code for a large refactoring project. </p>

<h3>Without AI Bridge: Rampant shadow AI</h3>

<p>A developer installs the CLI, generates a personal API key, and hardcodes database credentials so the agent can read the schema. If that laptop is compromised, the attacker has valid API keys and database access. The ops team has zero visibility into what code the agent generates or exports.</p>

<h3>With AI Bridge: Governed by default</h3>


Demo: Introduction to Coder's AI Bridge

<p>The developer opens their workspace and types <code>claude</code> in the terminal.</p>

<ul>
<li><strong>Immediate auth:</strong> The tool starts immediately without asking for login, using API credentials injected by AI Bridge.</li>
<li><a href="https://coder.com/docs/ai-coder/ai-bridge/mcp"><strong>Automatic MCP injection:</strong></a> When configured, AI Bridge can automatically inject all available tools from the configured MCP servers into each LLM call made by the agents and executes them on behalf of the user. This uses Coder&#39;s existing External Auth for authentication with remote MCP servers.</li>
<li><strong>Full audit:</strong> The security team sees: User <code>alice@example.com</code> used <code>150k</code> tokens and the <code>create_pull_request</code> GitHub MCP tool with the <code>claude-opus-4.5</code> model and <code>anthropic</code> provider with the prompt <code>create a PR with the proposed changes</code>.</li>
</ul>

<p>This transitions the agent from an isolated developer tool into a seamlessly integrated, friction-free part of the development workflow.</p>


<h2>Why this matters for your team</h2>

<p>For teams already running Coder, AI Bridge turns your development environments into a secure AI platform without forcing developers to change their workflow. For teams evaluating solutions, it&#39;s proof that you don&#39;t have to choose between developer velocity and security governance. </p>

<p>This architecture bridges the gap between <strong>Developer Experience</strong> (giving developers the best tools) and <strong>Security Governance</strong> (keeping the CISO happy) — a critical piece of any <strong>Secure AI Development Platform</strong>.</p>

<h3>For platform engineers</h3>

<p>It eliminates the &quot;help ticket&quot; loop of provisioning API keys for every new user. Configure the agents to use AI Bridge once in the Coder template, and it propagates to every workspace.</p>

<h3>For security teams</h3>

<p>It provides complete request logs listing every tool call, prompt, and token usage—partitioned by user, provider, and model. This is useful for compliance audits, incident response, and cost attribution. When a security incident is suspected, teams can quickly identify which user initiated which actions.</p>

<h3>For developers</h3>

<p>It means AI tools just work — no key management or waiting on tickets. The governance happens invisibly at the infrastructure level, so developers stay in flow.</p>

<h2>Ready to take control?</h2>

<p>Don&#39;t let shadow AI take over your engineering org. Centralize your keys, attribute users, and securely empower your agents.</p>

<p><a href="https://coder.com/docs/ai-coder/ai-bridge">Read the docs to get started.</a></p>


You don't have to choose between developer velocity and security governance. AI Bridge turns your development environments into a secure AI platform without forcing developers to change their workflow.

AI Bridge: How Coder Enables Observability and Control

<p>In software development, efficiency wins. Whether you’re automating the routine or accelerating delivery, every second counts. That’s why we’re excited to announce GitHub Copilot CLI is now available inside Coder workspaces and fully supported in Coder Tasks.</p>

<p>Copilot CLI brings natural language coding to the terminal. Built on Node.js and integrated with MCP, it supports advanced models like Claude Sonnet 4/4.5 and GPT-5. All you need is a GitHub Copilot subscription.</p>

<p>Let’s look at what this means inside Coder.</p>

<p>To get started, drop the following snippet to your existing Coder Template or check <a href="https://registry.coder.com/modules/coder-labs/copilot">Coder Registry</a> for detailed examples. </p>


<h2>Everyday dev tasks, now effortless</h2>

<p>Copilot CLI turns your terminal into a smarter, faster teammate. Here’s how it fits right into your flow:</p>

<p><strong>GitHub workflow automation</strong>  </p>

<p>Stay in your terminal. Move faster in GitHub.</p>

<pre><code>List my open PRs that need review
Create an issue for the authentication bug we just found
</code></pre>

<p>No tabs. No context switching. Just focus, right where you work.</p>

<p><strong>Upgrade dependencies and fix security issues</strong><br>
Copilot CLI helps you catch and correct vulnerabilities without breaking a sweat.</p>

<pre><code>Upgrade all dependencies with known vulnerabilities and run tests
</code></pre>

<p>It scans your project, updates your packages, manages breaking changes, and runs tests — all inside a Coder Task.</p>

<h2>Why GitHub Copilot CLI works better in Coder</h2>

<p>Copilot CLI is powerful on its own. But inside Coder, it becomes transformative:</p>

<p><strong>Integrated workspace experience</strong><br>
Everything runs securely inside your Coder workspace or Coder Task. No setup. No surprises.</p>

<p><strong>Natural language GitHub control</strong><br>
Skip the syntax. Use your words. Get what you need, fast.</p>

<p><strong>Smarter dependency management</strong><br>
Update packages. Handle breaking changes. Verify with tests — all on autopilot.</p>

<p><strong>Automated environment provisioning</strong><br>
From zero to ready, instantly. For anyone, on any team.</p>

<p><strong>Parallelize your work</strong><br>
With <a href="https://coder.com/docs/ai-coder/tasks">Coder Tasks</a>, run multiple copilot sessions in the background.</p>

<h2>Build smarter, move faster</h2>

<p>GitHub Copilot CLI helps developers stay productive. Inside Coder, it helps teams scale that productivity with structure, repeatability, and speed. Whether you’re reviewing pull requests, squashing bugs, or onboarding new hires, your Coder terminal now works as smart as you do.  </p>

<p>It’s another step toward a future where your dev environment doesn’t just run commands. It understands them.</p>


Add GitHub Copilot CLI with just a few lines of Terraform from the Registry, and integrate them with Coder.

Copilot CLI

GitHub Copilot CLI Now Runs Natively in Coder Workspaces

<p><em>April 2026 update: This post references Coder Tasks, an earlier approach to orchestrating agent workflows on Coder infrastructure. For the latest capabilities, check out <a href="https://coder.com/docs/ai-coder/agents">Coder Agents</a>.</em></p>

<hr>

<p>We are excited to announce the release of the <a href="https://docs.augmentcode.com/cli/overview">Auggie CLI agent</a> from Augment Code, now officially supported and ready for use in <strong>Coder Tasks!</strong> This integration lets you run Auggie directly inside your Coder workspace, so you can automate common tasks like dependency updates, code scaffolding, or environment setup without leaving your IDE or moving code out of your secure environment.</p>


<p>Integrating Auggie with Coder allows developers to leverage Augment Code’s powerful CLI agent within a secure and collaborative environment, offering unparalleled automation and efficiency directly within their workspaces</p>


<h2>What makes Auggie different from other CLI agents</h2>

<ol>
<li><p><strong>Deep codebase intelligence:</strong> Auggie doesn&#39;t just see the files you show it—it knows and understands your entire codebase autonomously using Augment&#39;s leading <a href="https://www.youtube.com/live/e7Wmnslt60M?feature=shared">context engine.</a></p></li>
<li><p><strong>Robust security features:</strong> Every action requires explicit proof-of-possession, enforced through SHA-256 file hashing. Service tokens verify organization membership and enforce scoped permissions, ensuring that access is always intentional and limited.  Even Auggie’s own engineers <a href="https://get.augmentcode.com/rs/926-RRL-702/images/Augment_Security_Principles.pdf">can’t bypass access controls</a>. They are SOC II Type 2, ISO 42001 certified. </p></li>
<li><p><strong>Created for the enterprise:</strong> Auggie was built for professional engineers and their teams working on large, complex codebases.</p></li>
</ol>


<h2>Auggie and Coder: Better together</h2>

<p>Pairing Auggie with Coder unlocks an extra layer of enterprise-grade control.</p>

<ul>
<li>With Auggie: you get a highly secure, cryptographically enforced agent, complete with proof-of-possession checks, scoped service tokens, and enterprise-grade certifications.</li>
<li>With Coder: you decide where that agent runs, whether that’s on your own infrastructure, in your private cloud, or even in an air-gapped environment. Unlike SaaS agent runners (such as Daytona, Cloudflare containers, or other hosted providers), Coder ensures your code and data never leave your compliance boundary.</li>
</ul>

<p>Together, you get the best of both worlds:</p>

<ul>
<li>Intelligent automation without sacrificing sovereignty: Auggie enforces what actions are allowed, while Coder ensures they only happen inside environments you control. This is especially powerful for regulated industries like finance, healthcare, and defense, where both agent-level security and execution-environment control are non-negotiable.</li>
<li>Seamless IDE collaboration: Auggie integrates directly with your preferred IDE inside Coder. Whether you use VS Code, IntelliJ, or another editor, Auggie can assist with automation tasks without breaking your flow.</li>
<li>Streamlined automation pipelines: Auggie’s autonomous agents can be orchestrated alongside your existing Coder Tasks, enabling a more automated development pipeline and freeing your team to focus on higher-level problem-solving and innovation.</li>
</ul>


<h2>Getting started with Auggie in Coder</h2>

<ol>
<li><a href="https://coder.com/docs/install">Install Coder</a></li>
<li>Set up your first <a href="https://coder.com/docs/admin/templates/creating-templates">template</a></li>
<li>Add <a href="https://registry.coder.com/modules/coder-labs/auggie">Auggie CLI module</a> by dropping the following snippet in your template.</li>
</ol>

<pre><code class="hcl">module &quot;auggie&quot; {
  source   = &quot;registry.coder.com/coder-labs/auggie/coder&quot;
  version  = &quot;0.1.0&quot;
  agent_id = coder_agent.example.id
  folder   = &quot;/home/coder/project&quot;
}
</code></pre>

<p>For any inquiries or support, please open an issue <a href="https://github.com/coder/registry">here</a> or <a href="https://discord.gg/coder">join</a> our Discord community.</p>

<p>With Auggie’s intelligent automation and Coder’s secure, developer-friendly environment, teams can confidently embrace autonomous agents without compromising security or compliance.</p>


The Auggie CLI agent from Augment Code is now officially supported and ready for use in Coder Tasks. This integration lets you run Auggie directly inside your Coder workspace, so you can automate common tasks like dependency updates, code scaffolding, or environment setup without leaving your IDE or moving code out of your secure environment.

Auggie CLI Agent Now Available in Coder Tasks

<p>With Coder&#39;s Cursor integration, template admins can configure Model Context Protocol (MCP) servers directly in template code, adding organization-specific context seamlessly. This happens without blocking the login process or disrupting existing developer workflows, ensuring smooth onboarding for teams adopting this solution. We implemented this to be reliable and highly scalable, accommodating growing team sizes while maintaining all configuration details under proper version control for IT governance and auditing, thereby addressing key compliance concerns for organizations.</p>


<h2>Why it matters</h2>

<ul>
<li><strong>Consistency across teams</strong> — All developer workspaces launch with identical MCP configurations across your organization without requiring any additional manual setup or intervention from individual team members. This eliminates environment-specific bugs and ensures that all developers work with the same tooling capabilities.</li>
<li><strong>Developer-first onboarding</strong> — New team members get access to the right tools and configurations the moment they open Cursor, eliminating the traditional learning curve and configuration headaches associated with AI tooling setup. This reduces time-to-productivity for new hires and allows them to contribute meaningful work from day one.</li>
<li><strong>Infra-as-code governance</strong> — All settings and configurations reside directly in Terraform alongside your other workspace templates, enabling systematic peer reviews, automated drift detection, comprehensive auditing capabilities, and straightforward rollbacks when needed. This integration with existing DevOps practices ensures that AI tooling configuration receives the same rigorous oversight as other infrastructure components.</li>
</ul>


<h2>How it works</h2>

<p>The module takes an optional <code>mcp</code> JSON input and uses <code>coder_script</code> during startup to write the configuration to <code>~/.cursor/mcp.json</code> in the developer environment. </p>

<p>The process runs seamlessly in the background during workspace startup without delaying the developer&#39;s connection experience. Cursor automatically detects and loads your MCP servers from the configuration file. This approach ensures developers can begin productive work immediately without needing any manual configuration steps.</p>


<h2>Example: Centrally Configure your MCP servers for Everyone</h2>


<p>Once the configuration is applied to your environment, every workspace becomes preconfigured to allow Cursor IDE to consume these MCP servers automatically upon connection—requiring no manual steps or intervention from developers. This automation eliminates the possibility of human error during configuration and ensures consistent implementation across all developer environments.</p>


<h2>Takeaway</h2>

<p>This update delivers enhanced repeatability, streamlined onboarding, and enterprise-grade governance for AI tooling in development environments. It speeds up time-to-code by eliminating manual setup, making it easy for teams to configure internal MCP servers. Your MCP servers become fully integrated into your infrastructure definition—properly versioned, systematically reviewed, and consistently deployed with your development workspaces.</p>


<h2>Getting Started</h2>

<ul>
<li><a href="https://registry.coder.com/modules/coder/cursor">Cursor module on the Coder Registry →</a></li>
<li><a href="https://docs.cursor.com/context/model-context-protocol">Cursor MCP configuration guide →</a></li>
</ul>


Enterprises are racing to adopt AI. Cursor IDE is most effective with Model Context Protocol (MCP). With Coder, you can pre-configure MCP in workspace templates so devs get a golden path from day one—while IT keeps control with IaC.

Coder Delivers Fast, Consistent MCP Tooling for Cursor

<p>In the last few months, we have been working on exciting new features for Coder. We are excited to share them with you today.</p>

<h2>Workspace Proxies (Experimental)</h2>

<p>Workspace proxies provide low-latency experiences for geo-distributed teams.</p>

<p>Coder&#39;s networking makes the best effort to connect directly to a workspace. In situations where this is not possible, such as developing with a web IDE (e.g. Jupyter), workspace proxies reduce latency for developers.</p>

<p>A workspace proxy is a relay connection a developer can use when connecting with their workspace over SSH, a workspace app, port forwarding, etc. Coder automatically selects the best proxy for the developer based on their location and the location of the workspace.</p>

<p>For deployment instructions, see <a href="https://coder.com/docs/v2/latest/admin/workspace-proxies">Docs</a>.</p>


<h2>Envbox</h2>

<p><a href="https://github.com/coder/envbox">Envbox</a> is an image developed and maintained by Coder that bundles the sysbox runtime. It works by starting an outer container that manages the various sysbox daemons and spawns an unprivileged inner container that acts as the user&#39;s workspace. The inner container is able to run system-level software similar to a regular virtual machine (e.g., systemd, dockerd, etc.). Envbox offers the following benefits over running sysbox directly on the nodes:</p>

<ul>
<li>No custom runtime installation or management on your Kubernetes nodes.</li>
<li>No limit to the number of pods that run Envbox.</li>
</ul>

<p>Refer to the <a href="https://coder.com/docs/v2/latest/templates/docker-in-workspaces#envbox">docs</a> for more information.</p>

<h2>Template Editor</h2>

<p>Edit &amp; duplicate templates directly from the Coder dashboard. This feature is available to all users with the <code>template-admin</code> or <code>owner</code> role.</p>


<h2>Fly.io Starter Template</h2>

<p>We have added a new starter template for <a href="https://fly.io/">Fly.io</a>. Fly.io is a platform for running applications globally. It is a great fit for Coder workspaces because it allows you to run them close to your developers. See the blog post <a href="https://coder.com/blog/remote-developer-environments-on-fly-io">here</a> for more information.</p>


<h2>Open in Coder</h2>

<p>Now you can generate <em>Open In Coder</em> buttons for your templates to provide a one-click experience for your developers. See the <a href="https://coder.com/docs/v2/latest/templates/open-in-coder">docs</a> for more information.</p>

<h3>Other Exciting Features</h3>

<ul>
<li>Ability to use non-localhost addresses with <code>coder port-forward</code> <a href="https://coder.com/docs/v2/latest/cli/port-forward">docs</a></li>
<li>Add or remove licenses from the dashboard.</li>
<li>Added support for X11 forwarding in the Coder CLI. <a href="https://coder.com/docs/v2/latest/cli/ssh">docs</a></li>
</ul>

<p>For a full chnagelog see our <a href="https://github.com/coder/coder/releases">releases</a> on GitHub.</p>


What's New in May 2023

<h1>Keep Coder templates up to date</h1>

<p>Coder <a href="https://coder.com/docs/v2/latest/templates">Templates</a> help keep each developer&#39;s workspace consistent. When a new template version is created (e.g. with a newer version of Java or more CPU cores), all developers are notified to update their workspaces.</p>

<p>It’s easy to update an existing template. You can make local changes to the template directly and then push the changes to your Coder server in multiple ways.</p>

<p>To keep your Coder templates up to date, you need to push the changes to Coder server. A user with a <code>template-admin</code> or <code>owner</code> role can update templates.</p>

<h2>Use the Coder CLI to push changes to the Coder template</h2>

<p>You can use the Coder CLI to push the changes to the template to the Coder server. This is the easiest way to update the template.</p>

<ol>
<li>Install the Coder CLI. See <a href="https://coder.com/docs/v2/latest/install">Installing the Coder CLI</a> for more information.</li>
<li><p>Push the changes to the Coder server:</p>

<pre><code class="shell">coder templates push &lt;template-name&gt;
</code></pre></li>
</ol>

<p>This will push the changes to the template to the Coder server, assign a random name to the template, and create a new template version. All users will then see the <strong>Update</strong> button on their workspace.</p>

<h2>Using CI/CD</h2>

<p>You can also use CI/CD to push the changes to the template to the Coder server. This is useful if you want to automate the process of updating the template. This is the preferred method if you have a version control system for your templates and want your developers to update the template by creating a pull request.</p>

<p>There are multiple ways to automate templates updates, like our <a href="https://coder.com/docs/reference/api">REST API</a>, <a href="https://coder.com/docs/reference/cli">CLI</a> and <a href="https://registry.terraform.io/providers/coder/coderd/latest"><code>terraform-provider-coderd</code></a>.
Here we describe how to push template updates using the <a href="https://github.com/marketplace/actions/setup-coder"><code>setup-coder</code></a> action which installs and authenticates the Coder CLI with your Coder server:</p>

<ol>
<li>Create a new repository for your template(s).</li>
<li>In your repository, create a new workflow called <code>push-to-coder.yml</code>. See <a href="https://docs.github.com/en/actions/learn-github-actions/introduction-to-github-actions#creating-a-workflow-file">Creating a workflow file</a> for more information.</li>
<li>Create a new personal access token with the <strong><code>template-admin</code></strong> or <strong><code>owner</code></strong> role. and add it to the repository action secrets with the name <code>CODER_SESSION_TOKEN</code>.</li>
<li><p>Use the <a href="https://github.com/marketplace/actions/setup-coder"><code>setup-coder</code></a> action from GitHub marketplace to create the workflow:</p>

<pre><code class="shell">coder tokens create --lifetime 8760h0m0s
</code></pre></li>
<li><p>Add the following steps to the <code>push-to-coder.yaml</code> workflow:</p>

<pre><code class="yaml">name: Push template to coder server

on:
    push:
        branches:
            - main
    pull_request:
        branches:
            - main
    workflow_dispatch:

jobs:
    deploy_template:
        runs-on: ubuntu-latest
        steps:
            - name: Checkout
                uses: actions/checkout@v4
                with:
                    submodules: true
            - name: Get short commit SHA # to use as template version name
                id: vars
                run: echo &quot;::set-output name=sha_short::$(git rev-parse --short HEAD)&quot;
            - name: Setup Coder
                uses: coder/setup-action@v1
                with:
                    access_url: &#39;https://coder.example.com&#39;
                    coder_session_token: ${{ secrets.CODER_SESSION_TOKEN }}

            - name: &quot;Push template&quot;
                run: |
                    coder templates push &lt;template-name&gt; /
                    --directory ./&lt;template-directory&gt; /
                    --name ${{ steps.vars.outputs.sha_short }} /
                    --yes
</code></pre></li>
<li><p>Push the changes to the Git repository. This will trigger the workflow and push the changes to the template to the Coder server.</p></li>
</ol>

<blockquote>
<p>See Coder&#39;s <a href="https://github.com/marketplace/actions/setup-coder"><code>setup-coder</code></a> on GitHub marketplace.</p>

<p>See Coder&#39;s <a href="https://github.com/coder/coder/blob/main/.github/workflows/dogfood.yaml">dogfood example workflow</a>.</p>

<p>See a community action <a href="https://github.com/marketplace/actions/update-coder-template">Update Coder Template</a> to automatically update templates.</p>
</blockquote>


Banner

Keep Coder Templates up to date

<p>Fly.io is a platform for deploying applications to the edge. It&#39;s a great fit for Coder because it&#39;s easy to deploy, scale, and manage. In this guide, we&#39;ll walk through deploying Coder on Fly.io and creating remote workspaces as Fly.io (firecracker) machines.</p>

<h2>Prerequisites</h2>

<ul>
<li>A Fly.io <a href="https://fly.io/signup">account</a>.</li>
<li><a href="https://fly.io/docs/getting-started/installing-flyctl/">flyctl</a> installed on your local machine.</li>
</ul>

<h2>Deploy Coder</h2>

<ol>
<li><p>Log in to Fly.io with the CLI:</p>

<pre><code class="bash">flyctl auth login
</code></pre></li>
<li><p>Create a new fly postgres database:</p>

<pre><code class="bash">flyctl postgres create --name &lt;postgres-app-name&gt;
</code></pre></li>
<li><p>Create a new fly app:</p>

<pre><code class="bash">flyctl apps create --name &lt;coder-app-name&gt;
</code></pre></li>
<li><p>Connect to the database with the coder fly app:</p>

<pre><code class="bash">flyctl postgres attach --app &lt;coder-app-name&gt; &lt;postgres-app-name&gt;
</code></pre>

<p><strong>Important</strong> Copy the connection string that is displayed in the terminal. You&#39;ll need it in the next step.</p></li>
<li><p>Create a new secret with the name <code>CODER_PG_CONNECTION_URL</code> and the value as the connection string you copied in the previous step.</p>

<pre><code>flyctl secrets set CODER_PG_CONNECTION_URL=&lt;connection-string&gt; --app &lt;coder-app-name&gt;
</code></pre></li>
<li><p>Edit the <code>fly.toml</code> file and update as per the example below:</p>

<pre><code class="toml">app = &quot;&lt;coder-app-name&gt;&quot; # Replace &lt;coder-app-name&gt; with the name of your app
kill_signal = &quot;SIGINT&quot;
kill_timeout = 5
primary_region = &quot;ams&quot;  # See a list of regions here: https://fly.io/docs/reference/regions/

[experimental]
    auto_rollback = true
    private_network = true  # Allows Coder to connect to the database

[build]
     image = &quot;ghcr.io/coder/coder:latest&quot;

[env]
    CODER_ACCESS_URL = &quot;https://&lt;coder-app-name&gt;.fly.dev&quot; # Replace &lt;app-name&gt; with the name of your app
    CODER_HTTP_ADDRESS = &quot;0.0.0.0:3000&quot;
    #CODER_VERBOSE = &quot;true&quot; # Uncomment this if you want to see more logs
    CODER_TELEMETRY_INSTALL_SOURCE = &quot;fly.io&quot;

[[services]]
    protocol = &quot;tcp&quot;
    internal_port = 3000
    processes = [&quot;app&quot;]

    [[services.ports]]
        port = 80
        handlers = [&quot;http&quot;]
        force_https = true

    [[services.ports]]
        port = 443
        handlers = [&quot;tls&quot;, &quot;http&quot;]
    [services.concurrency]
        type = &quot;connections&quot;
        hard_limit = 25

</code></pre></li>
<li><p>Deploy the app:</p>

<p>Run the following command to deploy the app from the directory where the <code>fly.toml</code> file is located:</p>

<pre><code class="bash">flyctl deploy --app &lt;coder-app-name&gt;
</code></pre></li>
<li><p>Scale the Coder app to 1GB of memory:</p>

<pre><code class="bash">flyctl scale memory 1024 --app &lt;coder-app-name&gt;
</code></pre></li>
<li><p>Congratulations! You&#39;ve deployed Coder on Fly.io!. Go to the URL of your app and create your first user!</p></li>
</ol>


<blockquote>
<p>If you want to use a custom domain, you can do so by following the instructions <a href="https://fly.io/docs/app-guides/custom-domains-with-fly/">here</a>.</p>
</blockquote>

<h2>Create your first template</h2>

<blockquote>
<p><strong>Warning</strong>: The following steps may not work as expected becasue Fly.io has deprecated and ended support of their terraform provider. See <a href="https://github.com/fly-apps/terraform-provider-fly">https://github.com/fly-apps/terraform-provider-fly</a> for more details.</p>
</blockquote>

<p>A template is a set of instructions that Coder uses to create a workspace. In this section, we&#39;ll create a template that uses a Fly.io machine as the workspace.</p>

<ol>
<li><p>Install Coder locally by following the instructions <a href="https://coder.com/docs/install">here</a> or if you are on linux/macOS, you can run the following command:</p>

<pre><code class="bash">curl -fsSL https://coder.com/install.sh | bash
</code></pre></li>
<li><p>Login to Coder:</p>

<pre><code class="bash">coder login &lt;coder-url&gt;
</code></pre></li>
<li><p>Create a new template by running the following command and following the prompts:</p>

<pre><code class="bash">coder templates init
</code></pre></li>
<li><p>Choose the <code>fly-docker-image</code> template and cd into the <code>fly-docker-image</code> directory.
<code>bash
cd fly-docker-image &amp;&amp; coder templates create
</code></p></li>
<li><p>Create the new template by running the following command from the <code>fly-docker-image</code> directory:</p>

<pre><code class="bash">coder templates push fly-docker-image --variable fly_api_token=$(flyctl auth token)
</code></pre></li>
</ol>


<p>If Coder is deployed as a fly.io app, instead of setting variable <code>fly_api_token</code> you can also set a fly.io secret with the name <code>FLY_API_TOKEN</code></p>

<pre><code class="bash">flyctl secrets set FLY_API_TOKEN=$(flyctl auth token)
</code></pre>

<p>Go to the Coder UI and create a new workspace using the <code>fly-docker-image</code> template.</p>


<p>This will deploy code-server on a Fly.io machine using the docker image. You can access the code-server instance by clicking on the <code>Code Server</code> button.</p>


<h2>Update Coder</h2>

<p>To update the Coder version, run <code>flyctl deploy --aap &lt;coder-app-name&gt;</code> again and it will pull the latest version of Coder.</p>

<h2>Next Steps</h2>

<p>To modify or write your first coder template, check out the <a href="https://coder.com/docs/v2/latest/templates">template docs</a>.</p>

<p>Have questions? <a href="https://coder.com/contact">Contact us</a> or <a href="https://discord.gg/coder">join us on Discord</a></p>


Remote developer environments on Fly.io

<p>Coder templates are the DNA used to create workspaces. They abstract the complexity of cloud environments, allowing developers to focus on their projects.</p>


Managing templates in Coder

<h2>Organizing your templates</h2>

<p>There are many ways to organize your templates. You can use them to:</p>

<ul>
<li><p><strong>Team based templates</strong>: A template for each of your teams e.g. a template for your <em>frontend</em> team and a template for your <em>backend</em> team.</p></li>
<li><p><strong>Project based templates</strong>: A template for each of your projects e.g. a template for your <em>cool-project</em> and a template for your <em>awesome-project</em>.</p></li>
<li><p><strong>Image based templates</strong>: A single template that is used for all your projects but with a different image for each project.</p></li>
</ul>


<h2>What is the difference between a template and an image?</h2>

<p>A template is a collection of settings that are used to create a workspace. An image is a collection of software that is used to create a workspace. A template can use one or more images. For example, you can have a template that uses the <em>ubuntu</em> image and the <em>node</em> image and the user will have the choice of which image to use when creating a workspace. Choices are managed by a terraform variable e.g.</p>

<pre><code class="hcl">    variable &quot;image&quot; {
      type        = string
      description = &quot;The image to use for the workspace&quot;
      default     = &quot;ubuntu&quot;
      validation {
        condition     = contains([&quot;ubuntu&quot;, &quot;node&quot;], var.image)
        error_message = &quot;The image must be either ubuntu or node&quot;
      }
    }
</code></pre>

<h2>Creating your first template</h2>


<ol>
<li><p>Create a new repository in your GitHub account. This will be the repository that contains your Coder templates.</p></li>
<li><p>Create a new directory in your repository called <code>deeplearning</code>.</p></li>
<li><p>Create a new file in the <code>deeplearning</code> directory called <code>main.tf</code>. This is the terraform file that will be used to create your template.</p></li>
</ol>

<pre><code class="hcl">   terraform {
     required_providers {
       coder = {
         source = &quot;coder/coder&quot;
         version = &quot;0.6.14&quot;
       }
       docker = {
         source = &quot;kreuzwerker/docker&quot;
         version = &quot;3.0.1&quot;
       }
     }
   }
   ...
</code></pre>

<ol>
<li>Create a rich-parameter variable of the form <code>data coder_parameter &lt;name&gt;</code>. This will be used to create a rich parameter in the template. For example:</li>
</ol>

<blockquote>
<p>See full list of available parameters <a href="https://registry.terraform.io/providers/coder/coder/latest/docs/data-sources/parameter">here</a></p>
</blockquote>

<pre><code class="hcl">    data &quot;coder_parameter&quot; &quot;framework&quot; {
      name        = &quot;Framework&quot;
      icon        = &quot;https://raw.githubusercontent.com/matifali/logos/main/memory.svg&quot;
      description = &quot;Choose your preffered framework&quot;
      type        = &quot;string&quot;
      mutable     = false
      default     = &quot;no-conda&quot;
      option {
        name        = &quot;PyTorch&quot;
        description = &quot;PyTorch&quot;
        value       = &quot;pytorch&quot;
        icon        = &quot;https://raw.githubusercontent.com/matifali/logos/main/pytorch.svg&quot;
      }
      option {
        name        = &quot;PyTorch Nightly&quot;
        description = &quot;PyTorch Nightly&quot;
        value       = &quot;pytorch-nightly&quot;
        icon        = &quot;https://raw.githubusercontent.com/matifali/logos/main/pytorch.svg&quot;
      }
      option {
        name        = &quot;Tensorflow&quot;
        description = &quot;Tensorflow&quot;
        value       = &quot;tensorflow&quot;
        icon        = &quot;https://raw.githubusercontent.com/matifali/logos/main/tensorflow.svg&quot;
      }
      option {
        name        = &quot;Tensorflow + PyTorch&quot;
        description = &quot;Tensorflow + PyTorch&quot;
        value       = &quot;no-conda&quot;
        icon        = &quot;https://raw.githubusercontent.com/matifali/logos/main/tf-torch.svg&quot;
      }
      option {
        name        = &quot;Tensorflow + PyTorch + conda&quot;
        description = &quot;Tensorflow + PyTorch + conda&quot;
        value       = &quot;conda&quot;
        icon        = &quot;https://raw.githubusercontent.com/matifali/logos/main/tf-torch-conda.svg&quot;
      }
      option {
        name        = &quot;Conda&quot;
        description = &quot;Only conda (install whatever you need)&quot;
        value       = &quot;conda-base&quot;
        icon        = &quot;https://raw.githubusercontent.com/matifali/logos/main/conda.svg&quot;
      }
    }
</code></pre>

<ol>
<li>For this example, we will use the pre-built <a href="https://github.com/matifali/dockerdl">DockerDL</a> images.</li>
</ol>

<blockquote>
<p><strong>Note</strong>: You can use any image you want. You can use a pre-built image or coder can build an image for you from the Dockerfile in <em>images</em> directory. See</p>
</blockquote>

<pre><code class="hcl">   data &quot;docker_registry_image&quot; &quot;dockerdl&quot; {
     name = &quot;matifali/dockerdl:${data.coder_parameter.framework.value}&quot;
   }
   resource &quot;docker_image&quot; &quot;dockerdl&quot; {
     name          = data.docker_registry_image.dockerdl.name
     pull_triggers = [data.docker_registry_image.dockerdl.sha256_digest]
     # Keep alive for other workspaces to use upon deletion
     keep_locally = true
   }
</code></pre>

<blockquote>
<p>Full example is available <a href="https://github.com/matifali/coder-templates/blob/main/deeplearning/main.tf">here</a></p>

<h2>Next steps</h2>
</blockquote>

<p>Have questions? <a href="https://coder.com/contact">Contact us</a> or <a href="https://discord.gg/coder">join us on Discord</a></p>

<p>Stay tuned for the next post: How to keep your templates up to date.</p>


Coder templates are the DNA used to create workspaces. They abstract the complexity of cloud environments, allowing developers to focus on their projects.

AI-native Development

Secure Environments for 2,500+ Developers

M Atif Ali

Explore the latest from M

AI Bridge: How Coder Enables Observability and Control

GitHub Copilot CLI Now Runs Natively in Coder Workspaces

Auggie CLI Agent Now Available in Coder Tasks

Coder Delivers Fast, Consistent MCP Tooling for Cursor

What's New in May 2023

Keep Coder Templates up to date

Remote developer environments on Fly.io

Managing templates in Coder

Related content

Using Claude Code with Coder Workspaces

Introducing Coder Agents

Self-Hosted Doesn’t Always Mean What It Implies